Capacity matters. So do location, cost and reliability. But for a data centre with an opening date to meet, the immediate challenge is often the gap between when usable power is needed and when it can reach the site.
That is the timing problem. A connection measured in megawatts is only useful to a development programme when its delivery date, operating conditions and available capacity match the plan. On-site generation can be worth evaluating where those dates diverge. It needs the same scrutiny as the grid connection itself.
A national ambition is not a site connection
The UK government’s Compute Roadmap forecasts a need for at least 6GW of AI-capable data-centre capacity by 2030. That is a forecast of required capacity, not a statement that the connections or facilities already exist. (UK Compute Roadmap, Action 6)
The scale of applications is larger still. In its July 2026 consultation, Ofgem put data-centre capacity in the demand connection queue at around 73GW. That figure describes queued capacity: it is neither actual electricity consumption nor a guarantee that every proposed development will be built. Ofgem identifies speculative projects as part of the problem. (Ofgem, July 2026 consultation)
Queue reform matters, but so does physical infrastructure. Ofgem identifies both non-viable projects blocking others and the time needed to build networks and generation as causes of slow connections. Improving the queue cannot remove every construction dependency. (Ofgem, June 2026 update)
For an individual development, the useful question is therefore specific: what power will be available at this location, on what date, and with what restrictions? A first phase might need less than the completed campus. Equally, a provisional date or restricted connection might leave a gap that the operating plan cannot absorb.
Put the opening date beside the power programme
Start with the load required at opening, then map how it grows. Include the IT equipment, cooling and other supporting systems. Distinguish the initial requirement from the eventual campus capacity, and identify what must remain available during maintenance or a failure.
Compare that demand profile with the connection offer and the work needed to make it usable. A credible programme should identify dependencies, evidence behind delivery dates and responsibility for each milestone. Delays can defer revenue and put commitments under pressure, but the commercial exposure belongs in the project’s own model.
This is where alternative supply arrangements become a practical assessment. Bridge generation is intended to cover a period before a grid connection becomes available. Longer-term on-site generation has a different operating life, fuel requirement and maintenance case. Neither should be selected simply because its equipment rating matches the headline demand.
The distinction also changes the exit plan. If generation is temporary, establish what happens when the grid arrives: removal, retention, a revised operating role or another agreed outcome. If it remains, assess how it will operate alongside the connection. The How It Works page provides a starting point for discussing that site assessment.
On-site power needs its own critical path
On-site generation is not automatically the faster option. The IEA’s 2026 analysis cautions that gas generation can face its own equipment constraints, particularly in turbine supply. A general market observation cannot establish the availability of a particular engine, supplier or package. (IEA, Key Questions on Energy and AI)
A realistic assessment needs confirmed equipment availability, fuel capacity and supply arrangements, the site layout, electrical integration and a commissioning programme. Planning and environmental requirements also belong at the beginning. Applicable combustion-plant and generator permits depend on the installation and jurisdiction; the headline electrical rating alone does not settle the question. (Government environmental permit guidance)
Factory assembly can move work away from the site, but it cannot remove these dependencies. A delivery date for equipment is not the same as a date for usable power.
Resilience needs separate attention. A plant intended to operate without the grid must be designed for that role, including appropriate controls, switching and starting arrangements. Maintenance, equipment failure and transitions between sources need to be addressed alongside the data centre’s UPS and other resilience systems. The total of the generator ratings is not a complete availability plan. (US Department of Energy, CHP resilience guidance)
Design electricity and cooling together
Generation also creates heat. Combined heat and power recovers useful heat alongside electricity; its value depends on matching that heat to a real demand. The Technology page introduces the generation and heat-recovery process. (Government CHP guidance)
In a suitable design, recovered engine heat can drive an absorption chiller to produce chilled water. This uses heat from generation; it should not be confused with recovering low-temperature heat from servers. Absorption cooling can reduce demand on electrically driven cooling, but it does not make cooling electricity-free. Pumps, controls and heat-rejection equipment still require power. (US Department of Energy, absorption chillers)
The amount and temperature of available heat, the required chilled-water conditions and the cooling demand determine whether the arrangement fits. Heat rejection also needs space and engineering: useful cooling does not mean the remaining heat disappears. Explore Cooling alongside the electrical design, rather than treating it as an efficiency claim to add afterwards.
The model should account for the whole installation and changing loads. It should not count recovered heat as a useful output and then count the cooling made from that same heat again. A credible assessment shows what energy enters, what useful services leave and what the supporting equipment consumes.
Make the carbon and commercial cases site-specific
Natural-gas generation produces emissions. The carbon comparison must use the electricity and cooling the proposed system would actually replace, realistic operating hours and the heat that can genuinely be used. Generic CHP efficiencies do not establish a particular data centre’s emissions reduction. Government guidance also stresses correct operation and maintenance when assessing performance. (Government CHP guidance)
The same discipline applies to cost. Fuel, maintenance, supporting electrical loads, connection works and the eventual role of the plant all belong in the assessment. The value of opening earlier is a separate project assumption, not a universal saving. The Sustainability page is a useful route into discussing the environmental questions alongside the operating case.
For further context, explore Deployments, then ask which evidence applies to the proposed configuration, operating conditions and delivery scope. An example from elsewhere is useful only when its differences are understood.
Start with the gap you need to close
The next step is to put the connection offer, opening load, growth plan and cooling requirements on one timeline. Assess grid-led, staged and on-site options against the same milestones and resilience requirements. The strongest proposal is the one whose assumptions can be checked and whose dependencies have owners.
Talk to an engineer about a site-specific power assessment. Bring the dates and loads that matter to the development. Those are the foundations for deciding whether on-site generation can close the gap—and what must be true for it to do so.